PRD Trace: a requirement-by-requirement check that an agent's PR did what the spec asked

The basic check, whether a PR satisfies its linked issue or spec, is already built into CodeRabbit, Qodo and Sourcery. Teams already pay for those tools, so a separate per-repo app that only does this would be hard to sell. The pre-commit CLI is in a worse position. Free Claude Code plugins and skills (spec-driven, Dev10x, several verification-before-completion skills, kspec) already block "done" until the acceptance criteria are verified. Charging for it would be difficult, and Anthropic or OpenAI could ship it as a built-in agent behavior. The remaining opening is narrow and mostly about depth and trust. The incumbents give an LLM verdict per issue or per criterion. I didn't see anyone splitting a long PRD into atomic requirements and linking each one to specific diff lines and to a specific test that covers it, ideally confirmed by running that test or checking coverage rather than by an LLM's opinion. An audit trail that persists across PRs would also be new: a requirement-to-code-to-test matrix that stays current as later PRs change the code. The buyers who need that kind of evidence are probably regulated or contract-driven teams in medical, fintech or agency work, not general small product teams. A new entrant would most likely win there, selling the persistent traceability record for audits and client sign-off. A PR comment alone is not enough of a product. These findings come from search-result summaries. I didn't check pricing, how mature these tools are, or how well their checks actually work.

• CodeRabbit (Assessment against linked issues)

  An established AI PR reviewer. It reads linked GitHub, GitLab, Jira or Linear issues, checks the PR against their acceptance criteria and flags gaps in an 'Assessment against linked issues' section. This covers the core of the GitHub app idea, although its output is an issue-level verdict, not a per-requirement table with line and test links.

• Qodo Spec Agent / requirement gaps

  Runs on every PR open or update, compares the implementation with the spec's requirements and reports requirement gaps as findings. It was labelled a research preview.

• Sourcery GitHub Issues integration

  Finds the issue most likely related to a PR, checks each part of the issue against the change and explains any gaps.

• PR Intent Checker

  A GitHub Action and MCP server that uses an LLM to compare a PR's diff with the requirements in the linked issue. It reports intent drift as PASS or FAIL.

• verify-pr skill (okisdev)

  A free agent skill. It parses Closes/Fixes #N, pulls the issue's acceptance criteria, checks each one against the diff and posts a verification table with pass/fail labels. This is close to the proposed table comment.

• spec-driven Claude Code plugin (uta2000)

  An MIT-licensed plugin whose Stop hook blocks the session until a verification agent has checked the acceptance criteria and returned PASS, FAIL or CANNOT_VERIFY with evidence. This is essentially the pre-commit trace idea, and it is free.

• Dev10x verify-acc-dod and other verification-before-completion skills

  Several open skills that make an agent verify acceptance criteria before a task is closed. Overlaps with the CLI idea.

• Tessl

  A spec-driven development platform. Specs link unit tests to requirements with [@test] links, and the agent checks that every requirement is satisfied when it finishes.

• kspec for Kiro

  An open-source CLI that adds 'kspec verify' to check that an implementation matches the spec in Kiro workflows.

• RTMX

  An open-source requirements traceability matrix toolkit for AI-driven development. It uses requirement IDs and has an MCP server so agents can query coverage.

• Zof

  A specification catalog that flags requirements with no mapping and tests with no requirement, which is traceability between requirements and tests.

Checked 2026-10-20